DuelLabGameBench 2
Main site
Benchmark menu

Archive

Archive

Historical benchmark records are kept here so the main Benchmark page can stay focused on GameBench 2 current standings. Archive pages preserve context without implying direct comparability.

Archive index

Benchmark lines

Compatibility

How to read old results

GameBench 1 remains useful as a historical record. GameBench 2 uses a different game set and evaluation context, so compare models within each benchmark rather than comparing ranks across them.

Immutable releases

Latest and milestone releases

This curated list highlights the latest package and releases that changed the benchmark line or its interpretation.

  • gb2-20260824-e05a1ff89a26 — ; latest package for data ID 20260824_103349.
  • gb2-20260806-59b3356f5989 — ; milestone release for data ID 20260806_094024. The August 6 standings introduced the v10 scoring policy and correction authority.
  • gb2-20260727-b63faa64fdbb — ; milestone release for data ID 20260727_140229. Reasoning Variants adopted model-relative mixed-variant scoring.
  • gb2-20260726-11fb0891c89b — ; milestone release for data ID 20260726_091516. Reasoning-bracket compatibility tracks and their source-bound policy were introduced.
  • gb2-20260720-959ae35a724b — ; milestone release for data ID 20260720_093741. Benchmark version 2-known8-r2 became the current eight-game line.
  • gb2-20260713-16054c869f49 — ; milestone release for data ID 20260713_194906. Initial GameBench 2 public release.

Complete canonical release history

Every additional canonical package remains directly discoverable here. Superseded citation aliases remain available in the complete machine-readable release lineage.